A rule‐based, feedback‐driven framework for fully automated prostate VMAT planning

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Abstract Background Manual prostate volumetric modulated arc therapy (VMAT) planning in Elekta Monaco is complex, time‐intensive, and prone to planner variability. Existing automation methods, such as knowledge‐based planning (KBP) and cloud‐based solutions often rely on external models rather than Monaco's native optimization. A treatment planning system (TPS)‐integrated, feedback‐driven approach could improve consistency and transparency while maintaining clinical quality. Purpose To develop and validate a fully automated planning algorithm (APA) that actively interacts with Monaco via its application programming interface (API), mimicking expert user behavior during optimization to deliver consistent, high‐quality plans. Methods The APA was developed as a rule‐based, fully deterministic workflow within the Monaco TPS, leveraging its native constrained and multicriterial optimization framework to enable end‐to‐end automated VMAT prostate plan generation. A retrospective, paired dosimetric study was conducted on localized prostate cancer cases. For each patient, the clinically approved manual plan was compared to an automated plan generated by the APA. The workflow reads target and organ‐at‐risk (OAR) objectives from the plan, monitors Monaco's optimization feedback—including dose‐volume histogram (DVH) metrics and cost‐function weights—and iteratively adjusts objectives. Endpoints included target coverage, OAR dose metrics, complexity, conformity, and planning time. Results Automated plans achieved non‐inferior target coverage with a median difference in PTV V57 Gy of −0.275% ( p = 0.116) between automated and manual plans. Automated plans produced statistically significant reductions in intermediate and low rectal doses including V40 Gy (1.62%, p = 0.003), V32 Gy (4.28%, p < 0.001), and V24 Gy (8.19%, p < 0.001), while high rectal doses (48–60 Gy) showed no statistically significant differences ( p > 0.05). Similarly for the bladder, automated plans significantly reduced V48 Gy (4.07%, p < 0.001) and V40 Gy (2.91%, p < 0.001), with a modest increase in high dose bladder volume (V60 Gy +1.33%, p < 0.001). Mean execution time for automated plans was 38.7 ± 14.7 min per case, compared to an informally estimated manual planning time of 2 h at our institution. All automated plans met institutional clinical acceptability criteria. Conclusions A TPS‐native, feedback‐driven automation using the Monaco Scripting API can replicate clinical planning strategies without human intervention, eliminating user variability and improving consistency. This approach offers a practical pathway for high‐quality prostate VMAT planning in routine clinical practice.